G TECHNO

IN L
G

OG
R
EME

IE
S
THAT

RLD
WO
WI

LL
E
T H
CHANGE

THE TECHNOLOGY REVIEW TEN

hat if you had a crystal ball that foretold the future of technology? Imagine, for example, if
W you had known in 1990 just how big the Internet was going to be 10 years hence. Sorry, that
crystal ball doesn’t exist. But in this special issue of Technology Review, we offer you the next best
thing: the educated predictions of our editors (made in consultation with some of technology’s top
experts). We have chosen 10 emerging areas of technology that will soon have a profound impact
on the economy and on how we live and work. These advances span information technology,
biotechnology and nanotechnology—the core of TR coverage in every issue. All of these areas merit
special attention in the decade to come. In each area we’ve chosen to highlight one innovator who
exemplifies the potential and promise of the field. Keep this issue around and see how well our pre-
dictions hold up—even without the aid of that crystal ball. —The Editors

TECHNOLOGY REVIEW January/February 2001 97

Duke University’s Miguel Nicolelis handles
a robotic arm. Brain signals from an owl
monkey (seen on the monitors to the right)
control the arm’s movement.

98 TECHNOLOGY REVIEW January/February 2001

 MIGUEL NICOLELIS
Brain-Machine
Interfaces
elle, a nocturnal owl monkey
B small enough to fit comfortably
in a coat pocket, blinks her out-
sized eyes as a technician plugs four con-
nectors into sockets installed in the top of
her skull. In the next room, measure-
ments of the electrical signals from some
90 neurons in Belle’s brain pulse across a
computer screen. Recorded from four
separate areas of Belle’s cerebral cortex,
the signals provide a window into what
her brain is doing as she reaches to touch
one of four assigned buttons to earn her
reward—a few drops of apple juice.
Miguel Nicolelis, a Duke University neu-
robiologist who is pioneering the use of
neural implants to study the brain, points
proudly to the captured data on the com-
puter monitor and says: “This readout is
one of a kind in the world.”
The same might be said of Nicolelis,
who is a leader in a competitive and high-
ly significant field. Only about a half-
dozen teams around the world are pur-
suing the same goals: gaining a better
understanding of how the mind works
and then using that knowledge to build
implant systems that would make brain
control of computers and other machines
possible. Nicolelis terms such systems
“hybrid brain-machine interfaces” or
HBMIs. Recently, working with the Lab-
oratory for Human and Machine Haptics
at MIT, he scored an important first on
the HBMI front, sending signals from
individual neurons in Belle’s brain to a
robot, which used the data to mimic the
monkey’s arm movements in real time.
In the long run, Nicolelis predicts
that HBMIs will allow human brains to
control artificial devices designed to
restore lost sensory and motor functions.
Paralysis sufferers, for example, might
gain control over a motorized wheel-
chair or a prosthetic arm—perhaps even
regain control over their own limbs.
“Imagine,” says Nicolelis, “if someone
could do for the brain what the pace-
maker did for the heart.” And, in much
the same way that a musician grows to
feel that her instrument is a part of her
own body, Nicolelis believes the brain
will prove capable of readily assimilating
human-made devices.
Ongoing experiments in other labs
Photograph by PAT R I C I A M C D O N O U G H
are showing that this idea is credible. At
Emory University, neurologist Phillip
Kennedy has helped severely paralyzed
people communicate via a brain implant
that allows them to move a cursor on a
computer screen (see “Mind Over Mus-
cles,” TR March/April 2000). And implants
may also shed light on some of the brain’s
unresolved mysteries. Nicolelis and other
neuroscientists still know relatively little
about how the electrical and chemical
signals emitted by the brain’s millions of
neurons let us perceive color and smell, or
give rise to the precise movements of
Brazilian soccer players—whose photos
adorn the walls of the São Paolo native’s
office. “We don’t have a finished model of
how the brain works,” says Nicolelis. “All
we have are first impressions.”
Others in Brain-Machine Interfaces
Organization Project
Andy Schwartz Neural control of robotic arm
(Arizona State University)
John Donoghue Brain representation of
(Brown University) movement
Richard Andersen Improved neuroelectrode
(Caltech) systems
Phillip Kennedy, Roy Bakay Communication systems for
(Emory University) paralyzed patients
Nicolelis’ latest experiments, how-
ever, show that by tapping into multiple
neurons in different parts of the brain, it
is possible to glean enough information
to get a general idea of what the brain is
up to. In Belle’s case, it’s enough infor-
mation to detect the monkey’s intention
of making a specific movement a few
tenths of a second before it actually hap-
pens. And it was Nicolelis’ team’s success
at reliably measuring tens of neurons
simultaneously over many months—pre-
viously a key technological barrier—that
enabled the remarkable demonstration
with the robot arm.
Still, numerous stumbling blocks
remain to be overcome before human
brains can interface reliably and comfort-
ably with artificial devices, making mind-
controlled prosthetic limbs or computers
more than just lab curiosities. Among the
key challenges is developing electrode
devices and surgical methods that will
allow safe, long-term recording of neu-
ronal activities. Nicolelis says he’s begun
working with Duke’s biomedical engi-
neering department to develop a teleme-
TECHNOLOGY REVIEW January/February 2001 99
try chip that would collect and transmit
data through the skull, without unwieldy
sockets and cables. And this year Nicolelis
will become co-director of Duke’s new
Center of Neuroengineering and Neuro-
computation, which will explore new com-
binations of computer science, chip design
and neuroscience. Nicolelis sees the effort
as part of an impending revolution that
could eventually make HBMIs as com-
monplace as Palm Pilots and spawn a
whole new industry—centered around
the brain. —Antonio Regalado
CHERIE KAGAN
Flexible Transistors
he implementation of pervasive
T computing—the spread of digi-
tal information throughout soci-
ety—will require electronics capable of
bringing information technology off the
desktop and out into the world (see
“Computing Goes Everywhere,” p. 52). To
digitize newspapers, product labels and
clothing, integrated circuits must be
cheap and flexible—a tough combination
for today’s silicon technology. Even the
cheapest form of silicon electronics—
the cut-rate “amorphous” silicon used
to drive laptop display screens—is too
pricey. What’s more, it’s difficult to incor-
porate silicon electronics on bendable
surfaces such as plastics.
Technology innovators are taking a
couple of routes around these limits.
Some researchers are trying to reinvent
amorphous silicon. Others have aban-
doned inorganic compounds like silicon
Others in Flexible Transistors
Organization Project
Lucent/Bell Labs Organic circuits
(Murray Hill, NJ)
Richard Friend Organic light-emitting diodes
(University of Cambridge)
Joseph Jacobson (MIT) Printed inorganics
Thomas Jackson Organic transistors
(Penn State)
to develop transistors based on organic
(carbon-based) molecules or polymers.
These organic electronics are inexpensive
to manufacture and compatible with
plastic substrates. Indeed, research teams
at places such as Lucent Technologies’
Bell Labs, England’s University of Cam-
100 TECHNOLOGY REVIEW January/February 2001
bridge and Pennsylvania State Universi-
ty have made impressive progress, and
commercial products are nearing the
market. Last fall, for example, Philips
Research in Eindhoven, the Netherlands,
showed off the first prototype of a rudi-
mentary display driven by polymer semi-
conductors. But there’s a catch: Organics
are far slower than their silicon cousins.
Now, a 31-year-old materials scientist
at IBM, Cherie Kagan, may have opened
the door to cheap, flexible electronics that
pack the mojo needed to bring ubiquitous
computing closer. Her breakthrough? A
compromise: transistors made from mate-
rials that combine the charge-shuttling
power and speed of inorganics with the
affordability and flexibility of organics.
These hybrids were created by
chemist David Mitzi at IBM’s Thomas J.
Watson Research Center in Yorktown
Heights, N.Y. By the time Kagan arrived
at Watson in 1998 following a stint at Bell
Labs (she earned a PhD from MIT in
1996), Mitzi had already shown that his
materials possessed intriguing electron-
ic properties. Kagan had a hunch they
might make good transistors. But she
needed quick results; she’d been hired
as a postdoc—a limited-time offer.
At the outset, the transistors flipped
on and off sluggishly. “The first times, I
IBM’s Cherie Kagan is making
transistors that could be far cheaper
and easier to fabricate than silicon
electronics. The reward: her own lab.
didn’t want to calculate [the speed],” says
Kagan. But she kept tweaking, and in less
than a year she had increased the mobil-
ity of electric charges through her tran-
sistors by four orders of magnitude—
matching the speed of amorphous silicon
and far exceeding most organic transis-
tors. The results won her a staff position
and her own lab at IBM.
Kagan has since increased the speed
by another 50 percent; further fine-tun-
ing, she believes, could provide at least
another doubling in acceleration. Not
only may the hybrids be far faster than
amorphous silicon, they have a key
advantage over silicon-based electron-
ics. Like some organic materials used to
make transistors, the hybrid materials
can be dissolved and printed onto paper
or plastic just like particles of ink. “I
make my materials in a different lab and
carry them over and add some liquid
and spin them on,” says Kagan. “It’s not
very sophisticated, which is sort of the
goal, right? You really want it to be cheap.”
Thomas Jackson, a transistor expert
at Penn State who is developing organic
circuits, says Kagan’s “fledgling results”
could pave the way for fast yet flexible
electronics. Jackson credits Kagan with
seizing the opportunity. “Not only does
she have her own pocket of competence,
Photograph by L A R R Y B U S A C C A
but she’s able to look around and see
exciting possibilities and then bring things
together. IBM has been working on these
sorts of materials for some time, but it
took the energy and enthusiasm and
vision and perspective of Cherie to trans-
late that into a thin-film transistor.”
The transistors could compete with
organic electronics in a variety of appli-
cations, such as radio-frequency product
ID tags. And then there’s the $20 billion-
per-year market for flat-panel video dis-
plays, where the speed of Kagan’s tran-
sistors could really make a difference.
Quicker circuits would deliver sharper
displays than those driven by amorphous
silicon at a fraction of the cost. That
would open the door to affordable wall-
sized displays or high-quality displays
that pop out of your pen. If all goes well,
the materials could be used in cheap,
flexible displays within five years.
Of course, bright displays that fit in
your pocket will require portable power,
and that has Kagan pondering her next
research challenge: cheap, flexible mate-
rials for solar cells to liberate pervasive
computing from bulky batteries. “You
aren’t going to want to carry a battery
around with your lightweight flexible
display,” she says. —Peter Fairley
DigiMine’s Usama Fayyad devises algorithms that detect meaningful patterns in massive collections of information.
Photograph by K A R E N M O S K O W I T Z
U S A M A FAY YA D
Data Mining
ello again, Sidney P. Manyclicks.
“
H We have recommendations for
you. Customers who bought this
title also bought...”
Intrusive? A touch of personal atten-
tion in the sterile world of e-shopping?
Both, perhaps—but definitely a tour de
force of database technology. Conven-
tional databases sort though a few
megabytes of structured data to find
answers to specific queries. But compil-
ing a simple recommendation list requires
a system that can burrow through giga-
bytes of Web site visitor logs in search of
patterns no one can anticipate in advance.
Welcome to data mining, also known
as knowledge discovery in databases
(KDD): the rapidly emerging technology
that lies behind the personalized Web
and much else besides. The emphasis
here is on “emerging,” says Usama Fayyad,
who should know: data mining didn’t
exist as a field until he helped pioneer it.
In 1987, the Tunisian-born comput-
er scientist was a graduate student at the
University of Michigan. He had taken a
summer job with General Motors, which
was compiling a huge database on car
TECHNOLOGY REVIEW
repairs. The idea, he says, was to enable
any GM service technician to ask the
database a question based on the model
of car, the engine capacity, and so on, and
get a quick, appropriate response. Sounds
straightforward. But, recalls Fayyad, “there
were hundreds of millions of records—no
human being could go through it all.” The
pattern recognition algorithm he devised
to solve that problem became his 1991
Others in Data Mining
Organization Project
Howard Wactlar Search very large video collections
(Carnegie Mellon)
Marti Hearst (University Automated discovery of new information
of California, Berkeley) from large text collections
Nokia Research Center Finding recurrent episodes in event
(Helsinki, Finland) sequence data
Raghu Ramakrishnan Visual exploration of data on the Web
(University of Wisconsin)
doctoral dissertation, which is still among
the most cited publications in the data-
mining field.
Data mining proved to have surpris-
ingly broad application. Fayyad left Michi-
gan for NASA’s Jet Propulsion Laboratory,
where he applied his techniques to astro-
nomical research. In particular, his algo-
January/February 2001 101
ithm helped in automatically determining
which of some two billion observed celes-
tial objects were stars and which were
galaxies. The tool also helped find volca-
noes on Venus from the huge number of
radar images being transmitted from space
probes. A geologist could retrieve the
image of a previously identified volcano;
the computer would then examine the
picture for patterns and search through
other images for similar patterns. That
worked so well, Fayyad says, that “pretty
soon the military intelligence people were
all over us, wanting to use it. And so were
doctors, who wanted to do automatic
searches of radiology images.” In 1995, in
response to this widening interest, Fayyad
and his colleagues planned a full-scale
international conference on KDD. The
conference drew about 500 participants,
more than double what had been expect-
ed. (KDD 2000 drew 950.)
By this time, with the Internet gush-
ing information onto everyone’s desk-
top, the urgency for data mining was
becoming evident in the corporate world.
IBM and other industry giants sensed a
market—and wanted in. Microsoft set
its sights on Fayyad and enticed him to
join the company’s research labs. “They
suggested that I take a look at databases
in the corporate world,” says Fayyad. “It
was pretty sad. In many companies, the
‘data warehouses’ were actually ‘data
tombs’: the data went in and were never
looked at again.” Fayyad joined Microsoft
in 1996 and organized a new research
group in data mining. “We looked at new
algorithms for scaling up to very large
databases—gigabytes or larger,” he says.
By decade’s end, Fayyad had caught
tions of free-form text documents. The
results are still preliminary, as various labs
experiment with natural-language pro-
cessing, statistical word counts and other
techniques. But the University of Califor-
nia at Berkeley’s LINDI system, to take one
example, has already been used to help
geneticists search the biomedical literature
and produce plausible hypotheses for the
function of newly discovered genes.
Another hot area, says Fayyad, is
“video mining”: using a combination of
speech recognition, image understanding
and natural-language processing tech-
niques to open up the world’s vast video
archives to efficient computer searching.
For instance, when Carnegie Mellon Uni-
versity’s Informedia II system is given
an archive of, say, CNN news clips, it
produces a computer-searchable index
by automatically dividing each clip into
individual scenes accompanied by tran-
scripts and headlines.
Fayyad hopes that ultimately the
techniques of data mining will become so
successful and so thoroughly integrated
into standard database systems that they
will no longer be thought of as exotic.
“People will just assume that their data-
base software will do what they need.”
—M. Mitchell Waldrop
RANJIT SINGH
Digital Rights
Management
itting in his office in McLean,
S Va., Ranjit Singh is at ground
zero of what may be the
biggest—and bloodiest—of the many
ContentGuard’s Ranjit Singh manages
access to online digital property.

the entrepreneurial bug sweeping through
computer science labs. “I realized that
even the organizations that loved the
idea of data mining were having trouble
just maintaining their data.” What they
needed, he reasoned, was a company to
host their databases for them, and provide
data-mining services on top of that. The
result was digiMine, a Kirkland, Wash.,
startup that opened for business in March
2000 with Fayyad as CEO.
And the future of data-mining tech-
nology? Wide open, says Fayyad—espe-
cially as researchers begin to move beyond
the field’s original focus on highly struc-
tured, relational databases. One very hot
area is “text data mining”: extracting unex-
pected relationships from huge collec-
battles that will shape the Internet during
the 21st century’s initial decade. The bat-
tle lines are sharply drawn. On one side
are owners of intellectual property, or
content: books, music, video, photo-
graphic images and more. On the other
are Internet users—think Napster—who
want content to be freely distributed.
And then there is Singh, president of
ContentGuard, a company that spun
out of research at Xerox’s Palo Alto
Research Center (PARC) on a mission to
commercialize content protection in a
wired world. “The Internet changes
everything,” says Singh, 48, an England-
born technology manager whose resume
glitters with senior positions at Xerox,
Citibank and Digital Equipment plus a
number of startups. “The Internet,”
Singh continues, “allows perfect repro-
duction of digital content and totally
frictionless distribution.” A few mouse
clicks sends a work to millions of users,
but the creators and owners of the con-
tent won’t necessarily collect dime one
(see “Your Work Is Mine!” TR Novem-
ber/December 2000).
Ouch! You can bet the pain felt by
content owners who see their stuff flying
everywhere via the Net will translate into
action. Which is what Singh and Con-
tentGuard are about. Digital rights man-
agement, or DRM, is “the catalyst for a
revolution in e-content,” says Singh.
“DRM will allow content owners to get
much wider and deeper distribution than

102 TECHNOLOGY REVIEW January/February 2001 Photograph by C H R I S H A R T L O V E


ever before,” he maintains. “You can see
who is passing your content to whom.”
Stripped to its essence, DRM—as
provided by ContentGuard and a number
of competitors—amounts to an encryp-
tion scheme with a built-in e-business
cash register. Content is encoded, and to
get the key a user needs to do some-
thing—maybe paying money, maybe pro-
viding an e-mail address. DRM providers
deliver the protection tools; it is up to con-
tent owners to set the conditions. Con-
tentGuard uses a “multiple key” approach
to content protection; anyone who
received bootleg content would have to
crack into it all over again. Thus, Singh
explains, “even if a hacker cracks into a
piece of content, he cannot distribute it.”
So why isn’t DRM ubiquitous? Two
reasons. First, content owners are in the
midst of a hard rethink about both pric-
ing and distribution. Suddenly they are
wrestling with issues of how to price
three listens to a song, say, or a download
of a low-resolution image that cannot
be forwarded to others. “Content owners
now are trying out different economic
models for valuing content,” says Singh,
whose company provides DRM tools to,
among others, John Wiley & Sons and
Houghton Mifflin. “DRM opens many
possibilities,” he adds.
The second issue may be the more
nettlesome: “The user experience has to
hide the complexity of the protection
technologies,” says Singh. Users need to be
TECHNOLOGY REVIEW
able to buy the content they want “with-
out needing special viewers or down-
loads and without putting the user
through hoops,” he argues. To resolve
that, ContentGuard has forged multiple
partnerships with digital standard-bear-
ers such as Microsoft and Adobe Sys-
tems, and has extended its technology
so that it applies across many media,
including books, music and video.
Captivating as the possibilities of
DRM are, it is still in its early days. Says
John Schwarz, CEO of Reciprocal, Inc., a
distributor of ContentGuard and other
DRM solutions: “We are probably a year
or so away from seeing broad adoption of
DRM by the marketplace.”
Some analysts are more skeptical:
Others in Digital Rights
Organization Project
InterTrust Technologies Develops peer-to-peer
(Santa Clara, Calif.) distributed DRM technology
Reciprocal DRM clearinghouse
(New York)
Digimarc Watermarking to embed an
(Tualatin, Ore.) imperceptible code
Alchemedia “Clever Content” platform
(San Francisco) safeguards digital content
“I’m not convinced content can be pro-
tected in the Internet era,” says Eric
Scheirer, who tracks DRM issues for For-
rester Research. “People want flexible
access to content.” Proof is Napster, of
course, which represents a phenomenon
Scheirer calls “unstoppable.” Even if Nap-
ster is put out of business by the courts,
he predicts that the frictionless distribu-
tion of digital content among the millions
of Internet users will live on.
But Singh is betting heavily that
DRM will prevail and, ironically, he
thanks Napster. “Napster turned this
whole issue into a CEO-level question.
The very highest corporate officers now
are looking into content management
issues, and they want to protect their
property.”
That, says Singh, augurs wider use of
DRM. “Here’s the virtual cycle you will
see: The more content a business puts
online, the faster it will want to put still
more content up, because it will see the
economic benefits and users will see the
benefits of gaining access to more con-
tent. That’s why we are seeing an explo-
sion here.” —Robert McGarvey
January/February 2001 103
J O S E P H AT I C K

Biometrics
n one sense, the field of biomet-

I rics—identifying individuals by
specific biological traits—has
already emerged. Large companies use
fingerprint sensors for logging on to cor-
porate networks, state driver’s license
authorities employ face recognition for
capturing and storing digital pho-
tographs, and the first iris-scan-protect-
ed ATM in the nation was introduced in
Texas in May 1999. Yet consumers have
been reluctant to adopt the technology,
and so far, it remains largely relegated to
military and government applications.
But the emergence of another tech-
nology—the wireless Web—could soon
change all that, according to Joseph Atick,
president and CEO of Visionics, one of the
leaders in face recognition technology.
“Personal digital assistants (PDAs) and cell
phones are becoming our portal to the
Visionics’ Joseph Atick sees the wireless
world, our transaction devices, our ID
Web as key to widespread consumer
and maybe one day our passport,” says
adoption of biometric technologies.
Atick. But entrusting these small gadgets
with so much of our personal and finan-
cial information carries with it a great skeptical of Atick’s vision of a biometric- to this process,” says Atick. In 1994, he and
risk.“It is this need for security,” Atick says, enabled wireless Web can’t deny his inge- colleagues Paul Griffin and Norman
“that is going to drive biometrics.” nuity and ambition. At the age of 15, Redlich founded Visionics.
And while the need for security is while living in Israel, Atick dropped out Based in Jersey City, N.J., Visionics
pushing the demand for biometric sys- of school to write a 600-page physics develops and markets pattern-recognition
tems, other technology developments— textbook entitled Introduction to Modern software called FaceIt. In contrast to the
increased bandwidth, new cell phones Physics. “I was bored in school. I wanted main competing technology, which relies
and handheld computers equipped with to show the establishment I was serious on data from the entire face, FaceIt veri-
digital cameras—will create an infra- about my interests,” says Atick. “This fies a person’s identity based on a set of
structure capable of putting biometrics book was my ticket to grad school.” four facial features that are unique to
into the hands of consumers. Visionics is Remarkably, Stanford University accept- the individual and unaffected by the pres-
taking advantage of this combination of ed him at 16 into its graduate program, ence of facial hair or changes in expres-
need and infrastructure by developing where he earned his master’s degree in sion. In the past few years, the system has
tools to enable people to authenticate physics and PhD in mathematical physics. found success fighting crime in England
any transaction they make over the wire- After graduation, Atick applied his and election fraud in Mexico.
less Web using their own faces. math skills to the study of the human In October, the company signed a
Even those in the industry who are mind. While heading the Computational merger agreement with Digital Biomet-
and Neuroscience Laboratory at Rocke- rics, a Minnetonka, Minn.-based bio-
feller University, he sought to understand metric systems engineering firm. Togeth-
Others in Biometrics
how the brain processes the abundance of er they plan to build the first line of
Organization Project visual information thrown at it by the “biometric network appliances”—com-
Visage Technology Face recognition environment. He and his colleagues dis- puters hooked to the Net with the capac-
(Littleton, Mass.) covered that the brain deals with visual ity to store and search large databases of
Iridian Technologies Iris recognition information much as computer algo- facial or other biometric information.
(Marlton, N.J.) rithms compress files. Because everyone The appliances, containing customers’
DigitalPersona Fingerprint recognition has two eyes, a nose and lips, the brain identification data, can then receive
(Redwood City, Calif.) extracts only those features that typically queries from companies wanting to
Cyber-SIGN Dynamic signature show deviations from the norm, such as authenticate e-transactions. And while
(San Jose, Calif.) verification the bridge of the nose or the upper cheek- consumers will be able to access the sys-
bones. The rest it fills in.“We soon realized tem from a cell phone, PDA or desktop
T-NETIX (Englewood, Colo.) Voice recognition
there was tremendous commercial value computer, Atick expects handheld devices

106 TECHNOLOGY REVIEW January/February 2001 Photograph by N I C K C A R D I L L I C C H I O

to be the biggest market. Visionics is also
working with companies in Japan and
Europe to have FaceIt software installed on
new Web-ready mobile devices so con-
sumers can capture their own faces and
submit encrypted versions over the Net.
Is that it for PINs and passwords?
Atick predicts it will still be two to three
years before PDA- and cell-phone-wield-
ing consumers are likely to use biometrics
instead. And as futuristic as his vision is,
he is really striving toward something a
bit old-fashioned. “Essentially, we are
bringing back an old element of human
commerce,” says Atick—restoring the
confidence that comes with doing busi-
ness face to face. —Alexandra Stikeman
KAREN JENSEN
Natural Language
Processing
he 1968 film 2001: A Space
T Odyssey gave us a vision of the
millennium based on the tech-
nological predictions of the day. One
result: HAL 9000, a computer that con-
versed easily with its shipmates like any
other crew member. The timing was off:
In the real 2001, there’s not a computer in
Photograph by K A R E N M O S K O W I T Z
the solar system as articulate as HAL.
But maybe it wasn’t that far off. HAL’s
modern-day counterparts are catching
up fast (sans the homicidal tendencies,
one hopes). Already we have commercial
speech recognition software that can take
dictation, speech generation equipment
that can give mute people voices and
software that can “understand” a plain-
English query well enough to extract the
right answers from a database.
Emerging from the laboratories,
moreover, is a new generation of inter-
faces that will allow us to engage com-
puters in extended conversation—an
activity that requires a dauntingly com-
plex integration of speech recognition,
natural-language understanding, dis-
course analysis, world knowledge, rea-
soning ability and speech generation. It’s
true that the existing prototypes can only
talk about such well-defined topics as
weather forecasts (MIT’s Jupiter), or local
movie schedules (Carnegie Mellon’s
Movieline). But the Defense Advanced
Research Projects Agency (DARPA) is
working on wide-ranging conversation-
al interfaces that will ultimately include
pointing, gesturing and other forms of
visual communication as well.
Parallel efforts are under way at
Microsoft Research’s Karen Jensen is
heading an effort to give machines the
ability to grok human language.
TECHNOLOGY REVIEW
industry giants such as IBM and
Microsoft, which see not only immediate
applications for computer users who
need to keep their hands and eyes free but
also the rapid evolution of speech-
enabled “intelligent environments.” The
day is coming when every object big
enough to hold a chip actually has one.
Others in Language Processing
Organization Project
Victor Zue (MIT Laboratory Conversational interfaces
for Computer Science)
Alexander I. Rudnicky Verbal interaction with small
(Carnegie Mellon) computers
Ronald A. Cole Domain-specific conversational
(University of Colorado) systems
BBN Technologies Dialog agent
(Cambridge, Mass.)
We’d better be able to talk to these objects
because very few of them will have room
for a keyboard.
Getting there will be a huge chal-
lenge—but that’s exactly what attracts
investigators like Karen Jensen, the gung-
ho chief of the Natural Language Pro-
cessing group at Microsoft Research. Says
Jensen: “I can’t imagine anything that
would be more thrilling, or carry more
potential for the future, than to make it
possible for us to truly interact with our
computers. That would be so exciting!”
Such declarations are typical of
Jensen, who at 62 remains as exuberant
about technology’s promise as any teenag-
er—and just as ready to keep hacker’s
hours. Indeed, Jensen was one of the first
people Microsoft hired when it opened its
research lab in 1991. Along with col-
leagues Stephen Richardson and George
Heidorn, she arrived at the Redmond,
Wash., campus from IBM’s Thomas J.
Watson Research Center, where they had
worked on some of the earliest gram-
mar-checking software, and immediate-
ly started building a group that now
numbers some 40 people.
In Redmond, Jensen and her col-
leagues soon found themselves con-
tributing to the natural-language query
interface for Microsoft’s Encarta ency-
clopedia and to the grammar checker
that first appeared in Word 97. And now,
she says, they’ve begun to focus all their
efforts on a unique technology known as
MindNet. MindNet is a system for auto-
matically extracting a massively hyper-
January/February 2001 107
linked web of concepts from, say, a
standard dictionary. If a dictionary
defines “motorist” as “a person who dri-
ves a car,” for example, MindNet will use
its automatic parsing technology to find
the definition’s underlying logical struc-
ture, identifying “motorist” as a kind of
person, and “drives” as a verb taking
motorist as a subject and car as an object.
The result is a conceptual network that
ties together all of human understanding
in words, says Jensen.
The very act of putting this concep-
tual network into a computer takes the
machine a long way toward “under-
standing” natural language. For example,
to figure out that “Please arrange for a
meeting with John at 11 o’clock” means
the same thing as “Make an appointment
with John at 11,” the computer simply has
Others in Microphotonics
Organization Project
Eli Yablonovitch Photonic crystals for optical
(UCLA) and radio frequencies
Susumu Noda Optical integrated circuits
(Kyoto University, Japan)
Axel Scherer Optical switches, waveguides
(Caltech) and lasers
Nanovation Technologies Integrated devices for telecom
(Miami)
Clarendon Photonics Filters for WDM
(Boston)
to parse the two sentences and show that
they both map to the same logical struc-
tures in MindNet. “It’s not perfect
grokking,” Jensen concedes. “But it’s a
darn good first step.”
MindNet also promises to be a pow-
erful tool for machine translation, Jensen
says. The idea is to have MindNet create
separate conceptual webs for English and
another language, Spanish, for example,
and then align the webs so that the Eng-
lish logical forms match their Spanish
equivalents. MindNet then annotates
these matched logical forms with data
from the English-Spanish translation
memory, so that translation can proceed
smoothly in either direction.
Indeed, says Jensen, who is now in the
process of passing on the leadership of the
group to the younger generation, Mind-
Net seems to tie together everything
they’ve been doing for the past nine years:
“All we see is doors opening. We don’t see
any closing!” —M. Mitchell Waldrop
108 TECHNOLOGY REVIEW January/February 2001
J O H N J OA N N O P O U LO S
Microphotonics
ight bounces off the small yellow
L square that MIT physics profes-
sor John Joannopoulos is show-
ing off. It looks like a scrap of metal,
something a child might pick up as a
plaything. But it isn’t a toy, and it isn’t
metal. Made of a few ultrathin layers of
non-conducting material, this photonic
crystal is the latest in a series of materials
that reflect various wavelengths of light
almost perfectly. Photonic crystals are
on the cutting edge of microphotonics:
technologies for directing light on a
microscopic scale that will make a major
impact on telecommunications.
In the short term, microphotonics
could break up the logjam caused by the
rocky union of fiber optics and electron-
ic switching in the telecommunications
backbone. Photons barreling through the
network’s optical core run into bottle-
necks when they must be converted into
the much slower streams of electrons
that are handled by electronic switches
and routers. To keep up with the Inter-
net’s exploding need for bandwidth, tech-
nologists want to replace electronic
switches with faster, miniature optical
devices, a transition that is already under
way (see “The Microphotonics Revolu-
tion,” TR July/August 2000).
Because of the large payoff—a much
faster, all-optical Internet—many com-
petitors are vying to create such devices.
Large telecom equipment makers, includ-
ing Lucent Technologies, Agilent Tech-
nologies and Nortel Networks, as well as
a number of startup companies, are
developing new optical switches and
devices. Their innovations include tiny
micromirrors, silicon waveguides, even
microscopic bubbles to better direct light.
But none of these fixes has the tech-
nical elegance and widespread utility of
photonic crystals. In Joannopoulos’ lab,
photonic crystals are providing the means
to create optical circuits and other small,
inexpensive, low-power devices that can
carry, route and process data at the speed
of light. “The trend is to make light do as
many things as possible,” Joannopoulos
says. “You may not replace electronics
completely, but you want to make light do
as much as you can.”
Conceived in the late 1980s, photon-
ic crystals are to photons what semicon-
ductors are to electrons, offering an excel-
lent medium for controlling the flow of
light. Like the doorman of an exclusive
club, the crystals admit or reflect specif-
ic photons depending on their wave-
length and the design of the crystal. In the
1990s, Joannopoulos suggested that
defects in the crystals’ regular structure
could bribe the doorman, providing an
effective and efficient method to trap the
light or route it through the crystal.
Since then, Joannopoulos has been a
pioneer in the field, writing the definitive
book on the subject in 1995: Photonic
Crystals: Molding the Flow of Light. “That’s
the way John thinks about it,” says MIT
materials scientist and collaborator Edwin
Thomas. “Molding the flow of light, by
confining light and figuring out ways to
make light do his bidding—bend, go
straight, split, come back together—in the
smallest possible space.”
Joannopoulos’ group has produced
several firsts. They explained how crys-
tal filters could pick out specific streams
of light from the flood of beams in wave-
length division multiplexing, or WDM,
a technology used to increase the
amount of data carried per fiber (see
“Wavelength Division Multiplexing,” TR
March/April 1999). The lab’s work on
two-dimensional photonic crystals set
the stage for the world’s smallest laser
and electromagnetic cavity, key com-
ponents in building integrated optical
circuits.
But even if the dream of an all-opti-
cal Internet comes to pass, another prob-
lem looms. So far, network designers
have found ingenious ways to pack more
and more information into fiber optics,
both by improving the fibers and by
using tricks like WDM. But within five to
10 years, some experts fear it won’t be
possible to squeeze any more data into
existing fiber optics.
The way around this may be a type of
photonic crystal recently created by
Joannopoulos’ group: a “perfect mirror”
that reflects specific wavelengths of light
from every angle with extraordinary effi-
ciency. Hollow fibers lined with this
reflector could carry up to 1,000 times
more data than current fiber optics—
offering a solution when glass fibers reach
their limits. And because it doesn’t absorb
and scatter light like glass, the invention
may also eliminate the expensive signal
amplifiers needed every 60 to 80 kilome-
 MIT’s John Joannopoulos is using photonic
crystals, like this two-dimensional version he’s
holding, to better control the flow of light.

Photograph by J O H N S O A R E S TECHNOLOGY REVIEW January/February 2001 109

 With his “aspect-oriented” approach, Xerox
PARC’s Gregor Kiczales is making computer
programs easier to write and maintain.

ters in today’s optical networks (see “Blaz- GREGOR KICZALES a rule, such as: “When adding a new func-
ing Data,” TR November/December 2000). tion to this application, always put a trace
Joannopoulos is now exploring the Untangling Code statement in.” Of course, the rule works
theoretical limits of photonic crystals. ity software engineers. With the only if people remember to follow it.
How much smaller can devices be made,
and how can they be integrated into opti-
cal chips for use in telecommunications
P touch of a button, their programs
let us make global fixes in a long
text, say, or a spreadsheet, yet program-
Other crosscutting capabilities include
security and synchronization—the abili-
ty to make sure that two users don’t try to
mers often need to correct their own work access the same data at the same time.
Others Untangling Code one tedious line at a time. That irony Both require programmers to write the
isn’t lost on Gregor Kiczales, principal same functionality into many different
Organization Project
scientist at Xerox’s Palo Alto Research areas of the application. Even a modest-
Mehmet Aksit Composition filters Center (PARC) and professor at the Uni- sized application can easily present 100
(University of Twente,
the Netherlands) versity of British Columbia in Vancou- crosscutting issues.
ver—and he has a fix in mind. Kiczales Programmers try to track these
Karl Lieberherr Adaptive programming
(Northeastern University) champions what he calls “aspect-oriented instances of repetition, so that when a
programming,” a technique that will allow capability needs to be changed or upgrad-
IBM Research HyperJ system for Java
(Yorktown Heights, N.Y.) programming
software writers to make the same kinds ed, it can be done uniformly throughout
of shortcuts that those of us in other pro- the program. But keeping track of cross-
Mira Mezini Enhancing modularity and fessions have been making for years. cutting concerns is an error-prone process.
(Univ. of Siegen, Germany) reusability of A-O software
One such “crosscutting” capability is Forget to upgrade just a few of these
logging—the ability to trace and record instances, and your code starts collecting
and, perhaps, ultrafast optical computers? every operation the application performs. bugs. “We’re forced to keep track of every-
Says Joannopoulos: “Once you start being Since any given command might touch thing in our heads,” says Kiczales.
able to play with light, a whole new world down on functionally unrelated areas of Kiczales’ proposed solution is to cre-
opens up.” —Erika Jonietz the code, programmers now must make ate a new category within a program-

110 TECHNOLOGY REVIEW January/February 2001 Photograph by M A R K G I L B E R T

ming language called an “aspect.” Aspects
allow programmers to write, view and
edit a crosscutting concern as a separate
entity. Once the programmer is happy
with it, a single keystroke will weave the
aspect into the code wherever it is need-
ed. It’s a smart, intuitive, neat solution to
an old problem. And what’s good for
programmers is good for the rest of us:
Widespread adoption of aspects holds
out the promise of less buggy upgrades,
shorter product cycles and, ultimately,
better and less expensive software.
The idea of aspects has been around
for many years and with many different
names. It’s called “adaptive programming”
at Northeastern University, “subjective
programming” at IBM, “composition fil-
tering” at the University of Twente in the
Netherlands and “multidimensional sep-
aration of concerns” elsewhere. But unlike
these other research projects, Kiczales and
his team at PARC have taken the concept
out of the lab and into the real world by
incorporating the idea of aspects into a
new extension of the programming lan-
guage Java. The beta version of this exten-
sion (called AspectJ) is available for free at
www.aspectj.org, and Kiczales plans to
make release 1.0 ready by June. “Major
changes in programming methodology
can take 30 years to gain widespread accep-
tance,” he says. Making aspects an exten-
sion to an existing standard should, he pre-
dicts, “cut the cycle by 15 or 20 years.”
While Kiczales admits the tools are
still a little raw, there are nevertheless
about 500 users of AspectJ today—most
of them finding existing tools inadequate
for creating long, complicated programs
in Java. Some have already found Aspec-
tJ so solid that they’ve used it in produc-
tion. One of these is Checkfree.com, a
company that makes software for auto-
matic bill payment. Checkfree sells both
C++ and Java versions of the software.
Rich Price, senior engineer, estimates
that AspectJ allowed his team to imple-
ment an important crosscutting capa-
bility in the Java version in four pro-
grammer-hours, whereas the C++ team,
with no aspect-oriented programming
tools at their disposal, took two pro-
grammer-weeks to do the same thing.
Using aspects, he says, “I make one
change, in one place, and it gets woven in
where it needs to be. I love that.”
By folding their ideas into a practical
Java extension, Kiczales and his team
Photograph by J O H N G O O D M A N
hope to make aspects part of the ver-
nacular of programming languages.
“AspectJ lets programmers work more
quickly and at a higher design level,” says
Kiczales. “We’ve learned that crosscutting
concerns are actually not hard to work
with—once you have the proper pro-
gramming support.” —Claire Tristram
JORDAN POLLACK
Robot Design
obot builders make a convinc-
R ing case that in 2001, robots are
where personal computers were
in 1980—poised to break into the mar-
ketplace as common corporate tools and
ubiquitous consumer products perform-
ing life’s tedious chores. One big obstacle
remains: It is expensive to design and
make robots smart enough to adapt read-
ily to different tasks and physical envi-
ronments, the way human beings do.
That’s the reason why robotics have,
so far, found a commercial niche only in
simple and highly repetitive jobs, such as
working on an automotive assembly line,
or mass-producing identical items, such
as toys. The challenge for builders of
robots is to build more complexity into
them without the huge investment of
custom-tailoring each robot for a differ-
ent task.
One promising approach is to fully
Brandeis’ Jordan Pollack with some of his “creatures” designed by robots.
TECHNOLOGY REVIEW
automate the design and manufacture of
robotics by deploying computers to con-
ceive, test and even build the configura-
tions of each robotic system: in short, to
use robots to build robots. Last year, in a
cramped lab at Brandeis University in
Waltham, Mass., Jordan Pollack demon-
strated how this automated robotic design
and manufacturing might work.
Pollack, an associate professor of com-
puter science, together with postdoc Hod
Lipson, directed a computer to design a
moving creature using a limited set of
simple parts: plastic rods, ball joints, small
motors and a “brain” (neural network).
The computer—using an algorithm
inspired by biological evolution—
“evolved” hundreds of generations of
potential designs, killing off the sluggish
and refining the strong. Eventually, several
of the fastest and fittest came to life,
Others in Robot Design
Organization Project
Sarcos Robots for industry, medicine,
(Salt Lake City, Utah) Hollywood
iRobot Household communications
(Somerville, Mass.) robot
Humanoid Interaction Lab Interactions between humanoid
(Tsukuba, Japan) robots and humans
MIT Artificial Intelligence Machine learning, robot legs,
Lab (Cambridge, Mass.) faces
Robotics Institute Mobile robots and face
(Carnegie Mellon) recognition
January/February 2001 111
manufactured in a rapid-prototyping
machine. Pollack and Lipson snapped
on the motors, and the creatures moved.
“I think the important point of our
coevolutionary design and automated
manufacturing for robotics is to get small-
quantity production to be economical,”
Pollack says. He predicts that the evolu-
tionary approach to robot building could
lead to the first cheap industrial robots in
five to 10 years. “If we are successful, we
could see an industry within a decade
which makes low-quantity custom ma-
chinery worth more than it costs to make.”
For now, Pollack’s “automated”
process still takes plenty of human inter-
vention and money: Pollack and his col-
league wrote the computer program and
spent $50,000 on the human-built fabri-
cating device. Still, the team’s advance,
reported last August in the journal
Nature, garnered wide publicity. “The
importance is symbolic,” says Hans
Moravec, principal research scientist at
the Robotics Institute at Carnegie Mellon
University in Pittsburgh. “You have sys-
tems that develop robots out of thin air,
not by humans. In the future, there will be
real robots designed that way.”
Pollack’s design and manufacturing
methods have plenty of competition.
Academic and industrial labs around the
world are busy building new generations
Others in Microfluidics
Organization Project
Aclara BioSciences Genomics and drug screening
(Mountain View, Calif.)
Caliper Technologies DNA, RNA and protein assays
(Mountain View, Calif.)
Cepheid DNA analysis
(Sunnyvale, Calif.)
Micronics Diagnostics and chemical
(Redmond,Wash.) analysis
TECAN Drug discovery
(Hombrechtikon, Switz.)
of robots. Within this decade, experts
predict a steady evolution in commercial
utility robots: robots that can clean floors
and pick up things. “There will be a mass
market for robots,” suggests George
Bekey, founder of the robotics lab at the
University of Southern California in Los
Angeles. “This next decade will be the
decade of the robot.”
Before robots reach out into the
everyday world of business and the house-
112 TECHNOLOGY REVIEW January/February 2001
hold, though, they will need their own ver-
sion of Moore’s Law: becoming dramat-
ically more affordable and powerful over
time. In spite of intriguing experiments
such as Pollack’s, designing even relative-
ly simple robots is a painstaking task. In
Japan, for example, Honda has spent over
14 years building a humanoid robot able
to walk, open a door and navigate stairs.
A walk around Pollack’s lab suggests,
perhaps, a better way to design robots. On
a workbench sits one example of his
computer-designed and computer-build-
able machines; it moves eerily like an
inchworm. Pollack trims excess plastic
from a newly fabricated plastic-rod
machine, oblivious to the shavings col-
lecting on his shirt and around his chair.
In a few years Pollack may well evolve a
cheap robot able to sweep those shavings
off the floor. —David Talbot
STEPHEN QUAKE
Microfluidics
he forces of physics move oceans,
T mountains and galaxies. But
applied physicist Stephen Quake
uses them to manipulate things on a vast-
ly reduced scale: tiny volumes of fluids
thousands of times smaller than a dew-
drop. Microfluidics, as Quake’s field is
called, is a promising new branch of
biotechnology. The idea is that once you
master fluids at the microscale, you can
automate key experiments for genomics
and pharmaceutical development, per-
form instant diagnostic tests, even build
implantable drug-delivery devices—all
on mass-produced chips. It’s a vision so
compelling that many industry observers
predict microfluidics will do for biotech
what the transistor did for electronics.
Quake’s 11-person lab at Caltech is
not the only outfit bent on realizing this
vision. Over the past decade or so, scores
of researchers have set out to build
microscale devices for many of the basic
processes of biological research, from
sample mixing to DNA sequencing. But
many of those groups have run into road-
blocks in developing technology that can
be generalized to a broad range of appli-
cations and would allow several func-
tions—such as sample preparation, DNA
extraction and detection of a gene muta-
tion—to be integrated on a single chip.
Moreover, some of the manufacturing
approaches involved, particularly silicon
micromachining, are so expensive that
experts in the field question whether
products relying on these techniques
could ever be economical to manufacture.
Quake’s group is one of several now
working their way around these obstacles.
Last spring, the team unveiled a set of
microfabricated valves and pumps—a
critical first step in developing technolo-
gy general enough to work for any
microfluidics application. And to make
microfluidic devices cheaper, Quake and
others are casting them out of soft silicone
rubber in reusable molds, using a tech-
nique called “soft lithography.” The
potential payoff of these advances is huge:
mass-produced, disposable microfluidic
chips that make possible everything from
drug discovery on a massive scale to at-
home tests for common infections.
Because microfluidics is so promising
and yet so technically frustrating, expec-
tation and hype have sometimes out-
paced the development of viable tech-
nology. Yet Quake and his group have
consistently turned out elegant devices
that actually work. First was a microscale
DNA analyzer that operates faster and on
different principles than the conven-
tional, full-sized version, then a miniature
cell sorter and most recently, those valves
and pumps, described last April in the
journal Science. All this while regularly
publishing important findings on the
basic physics of biological molecules.
If Quake seems adept at straddling
fields—in this case science and technol-
ogy—perhaps it’s because that’s exactly
the sort of challenge he has long craved.
Even as an undergraduate at Stanford
University, where he earned bachelor’s
and master’s degrees simultaneously in
only four years, Quake worried that
physics was “somewhat finished” as an
experimental science, that it was hard to
find the field’s frontiers. A pioneer at
heart, Quake started looking to tackle
questions that lay at the boundaries
between disciplines. As he recalls: “It was
completely obvious, even to an outsider,
that biology was going through this peri-
od of incredible growth and intellectual
excitement, and there were going to be big
questions asked and answered, and the
frontiers were advancing at a tremen-
dous rate in all directions.”
After Quake finished his doctorate in
theoretical physics at Oxford University,
 Caltech’s Stephen Quake has set his
sights on the microscale, building
tiny disposable devices that could
revolutionize biotechnology.
he went back to Stanford as a fellow
working on the physics of DNA. When
Caltech’s applied physics department
hired him in 1996, Quake says, “it was an
experiment for them”—he was the first
faculty member in the department with
a biological bent. So far, the experiment
seems to be going smoothly; this past
summer, at only 31, Quake got tenure.
Quake’s inventions are also thriving
in industry, through a startup called
Mycometrix. Founded in 1999 by Quake,
two of his college classmates and a con-
sultant, the South San Francisco-based
company has licensed all of Quake’s
microfluidics patents from Caltech. When
Photograph by M I C H A E L G R E C C O
TR went to press, the company was plan-
ning to deliver its first microfluidic
devices to selected university researchers
and industry partners by the end of 2000,
and was hoping for a commercial release
by the end of this year or early 2002. The
competition will be intense. Several star-
tups and even electronics giants like
Hewlett-Packard and Motorola are get-
ting in on the game. But to date, only one
of Mycometrix’s competitors has brought
a microfluidic product to market.
Although Quake’s work is rapidly
flowing into the commercial market-
place, it’s still the very early stages of sci-
ence and technology development that
TECHNOLOGY REVIEW
interest him the most. And though he has
built quite a reputation as a technologist,
he hopes soon to focus more of his atten-
tion on some of the most pressing ques-
tions in basic biology: How do the pro-
teins that control gene expression work?
How can you do studies that cut across
the entire genome? “Now that we’ve got
some pretty neat tools,” Quake says, “we’re
going to try and do some science with
them.” Quake’s ability to work in areas
from basic research to hot commercial
markets make him a prototypical inno-
vator. And the same versatility makes
microfluidics a field to pay close attention
to in the next few years.—Rebecca Zacks
January/February 2001 113